Mod: Update forward to compute all audio files per batch and improve UI for hyperparameters.
Browse files
app.py
CHANGED
@@ -23,68 +23,92 @@ def main() -> None:
|
|
23 |
|
24 |
model = load_conette(model_kwds=dict(device="cpu"))
|
25 |
|
26 |
-
|
27 |
-
allow_rep_mode = st.selectbox("Allow repetition of words", ["stopwords", "all", "none"], 0)
|
28 |
-
beam_size: int = st.select_slider( # type: ignore
|
29 |
-
"Beam size",
|
30 |
-
list(range(1, 21)),
|
31 |
-
model.config.beam_size,
|
32 |
-
)
|
33 |
-
min_pred_size: int = st.select_slider( # type: ignore
|
34 |
-
"Minimal number of words",
|
35 |
-
list(range(1, 31)),
|
36 |
-
model.config.min_pred_size,
|
37 |
-
)
|
38 |
-
max_pred_size: int = st.select_slider( # type: ignore
|
39 |
-
"Maximal number of words",
|
40 |
-
list(range(1, 31)),
|
41 |
-
model.config.max_pred_size,
|
42 |
-
)
|
43 |
-
|
44 |
-
st.markdown("Recommanded audio: lasting from **1 to 30s**, sampled at **32 kHz**.")
|
45 |
audios = st.file_uploader(
|
46 |
-
"Upload
|
47 |
type=["wav", "flac", "mp3", "ogg", "avi"],
|
48 |
accept_multiple_files=True,
|
49 |
)
|
50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
if audios is not None and len(audios) > 0:
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
)
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
|
89 |
|
90 |
if __name__ == "__main__":
|
|
|
23 |
|
24 |
model = load_conette(model_kwds=dict(device="cpu"))
|
25 |
|
26 |
+
st.warning("Recommanded audio: lasting from **1 to 30s**, sampled at **32 kHz**.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
audios = st.file_uploader(
|
28 |
+
"Upload audio files here:",
|
29 |
type=["wav", "flac", "mp3", "ogg", "avi"],
|
30 |
accept_multiple_files=True,
|
31 |
)
|
32 |
|
33 |
+
with st.expander("Model hyperparameters"):
|
34 |
+
task = st.selectbox("Task embedding input", model.tasks, 0)
|
35 |
+
allow_rep_mode = st.selectbox(
|
36 |
+
"Allow repetition of words", ["stopwords", "all", "none"], 0
|
37 |
+
)
|
38 |
+
beam_size: int = st.select_slider( # type: ignore
|
39 |
+
"Beam size",
|
40 |
+
list(range(1, 21)),
|
41 |
+
model.config.beam_size,
|
42 |
+
)
|
43 |
+
min_pred_size: int = st.select_slider( # type: ignore
|
44 |
+
"Minimal number of words",
|
45 |
+
list(range(1, 31)),
|
46 |
+
model.config.min_pred_size,
|
47 |
+
)
|
48 |
+
max_pred_size: int = st.select_slider( # type: ignore
|
49 |
+
"Maximal number of words",
|
50 |
+
list(range(1, 31)),
|
51 |
+
model.config.max_pred_size,
|
52 |
+
)
|
53 |
+
|
54 |
+
if allow_rep_mode == "all":
|
55 |
+
forbid_rep_mode = "none"
|
56 |
+
elif allow_rep_mode == "none":
|
57 |
+
forbid_rep_mode = "all"
|
58 |
+
elif allow_rep_mode == "stopwords":
|
59 |
+
forbid_rep_mode = "content_words"
|
60 |
+
else:
|
61 |
+
ALLOW_REP_MODES = ("all", "none", "stopwords")
|
62 |
+
raise ValueError(
|
63 |
+
f"Unknown option {allow_rep_mode=}. (expected one of {ALLOW_REP_MODES})"
|
64 |
+
)
|
65 |
+
del allow_rep_mode
|
66 |
+
|
67 |
+
kwargs: dict[str, Any] = dict(
|
68 |
+
task=task,
|
69 |
+
beam_size=beam_size,
|
70 |
+
min_pred_size=min_pred_size,
|
71 |
+
max_pred_size=max_pred_size,
|
72 |
+
forbid_rep_mode=forbid_rep_mode,
|
73 |
+
)
|
74 |
+
|
75 |
if audios is not None and len(audios) > 0:
|
76 |
+
audio_to_predict = []
|
77 |
+
cands = [""] * len(audios)
|
78 |
+
tmp_files = []
|
79 |
+
tmp_fpaths = []
|
80 |
+
audio_fnames = []
|
81 |
+
|
82 |
+
for i, audio in enumerate(audios):
|
83 |
+
audio_fname = audio.name
|
84 |
+
audio_fnames.append(audio_fname)
|
85 |
+
cand_key = f"{audio_fname}-{kwargs}"
|
86 |
+
|
87 |
+
if cand_key in st.session_state:
|
88 |
+
cand = st.session_state[cand_key]
|
89 |
+
cands[i] = cand
|
90 |
+
else:
|
91 |
+
tmp_file = NamedTemporaryFile()
|
92 |
+
tmp_file.write(audio.getvalue())
|
93 |
+
tmp_files.append(tmp_file)
|
94 |
+
audio_to_predict.append((i, cand_key, tmp_file))
|
95 |
+
|
96 |
+
tmp_fpath = tmp_file.name
|
97 |
+
tmp_fpaths.append(tmp_fpath)
|
98 |
+
|
99 |
+
if len(tmp_fpaths) > 0:
|
100 |
+
outputs = model(
|
101 |
+
tmp_fpaths,
|
102 |
+
**kwargs,
|
103 |
+
)
|
104 |
+
for i, (j, cand_key, tmp_file) in enumerate(audio_to_predict):
|
105 |
+
cand = outputs["cands"][i]
|
106 |
+
cands[j] = cand
|
107 |
+
st.session_state[cand_key] = cand
|
108 |
+
tmp_file.close()
|
109 |
+
|
110 |
+
for audio_fname, cand in zip(audio_fnames, cands):
|
111 |
+
st.success(f"**Output for {audio_fname}:**\n- {format_cand(cand)}")
|
112 |
|
113 |
|
114 |
if __name__ == "__main__":
|